Strong Singularity of Singular Masas in II_1 Factors

0.13 MB·English

by  Allan Sinclair

STRONG SINGULARITY OF SINGULAR MASAS
IN II FACTORS
1
Allan M. Sinclair Roger R. Smith∗
6
0
Stuart A. White Alan Wiggins
0
2

n School of Mathematics Department of Mathematics
a
University of Edinburgh Texas A&M University
J
Edinburgh EH9 3JZ College Station, TX 77843
4
2 Scotland USA

[email protected] [email protected]
]
A
[email protected] [email protected]
O
.
h
t
a
m
[
Abstract

1
v
4 A singular masa A in a II1 factor N is defined by the property that any unitary
9 w ∈ N for which A = wAw∗ must lie in A. A strongly singular masa A is one that
5
1 satisfies the inequality
0
6
0 kEA−EwAw∗k∞,2 ≥ kw−EA(w)k2
/
h
t for all unitaries w ∈ N, where EA is the conditional expectation of N onto A, and
a
m k·k∞,2 is defined for bounded maps φ : N → N by sup{kφ(x)k2 : x ∈ N, kxk ≤ 1}.
: Strong singularity easily implies singularity, and the main result of this paper shows
v
Xi the reverse implication.
r
a
∗Partially supported by a grant from the National Science Foundation.
1 Introduction
In [13], the first two authors introduced the concept of a strongly singular maximal
abelian self–adjoint subalgebra (masa) in a II1 factor N. For a bounded map φ: M →
N betweenanytwofinitevonNeumannalgebraswithspecifiedtraces, thek·k∞,2–norm
is defined by
kφk∞,2 = sup{kφ(x)k2 :x∈ M, kxk ≤ 1}. (1.1)
A masa A ⊆ N is then said to be strongly singular when the inequality
kEA−EwAw∗k∞,2 ≥ kw−EA(w)k2 (1.2)
holds for all unitaries w ∈ N, where the notation E indicates the unique trace pre-
B
servingconditional expectation ontoavon NeumannsubalgebraB. Anyunitarywhich
normalizes A is forced, by this relation, to lie in A, and so A is singular, as defined
in [3]. The original purpose for introducing strong singularity was to have a metric
condition which would imply singularity, and which would be easy to verify in a wide
range of cases (see [13, 14] and the work of the third author, [15], on Tauer masas
in the hyperfinite II1 factor R). Subsequently, [11], it was shown that every singular
masa in a separable II1 factor (where this terminology indicates norm–separability of
the predual N ) satisfies the following weaker inequality, analogous to (1.2):
∗
90kEA −EwAw∗k∞,2 ≥kw−EA(w)k2 (1.3)
holds for all unitaries w ∈ N. This clearly suggested that every singular masa should
be strongly singular, and our objective in this paper is to prove this result.
In 1983 Sorin Popa introduced the δ–invariant for a masa in a II1 factor, [7]. This
was the first attempt to define a metric based invariant for a masa, which he used
to show that there is an abundance of singular masas in separable II1 factors, [7].
Subsequently Popa, [8], showed that a masa in a separable II1 factor is singular if, and
only if, it has δ–invariant 1. An invariant α(A) for a masa A in a II1 factor, based
on unitary perturbations of A, was defined by the first two authors in [13] with strong
singularity corresponding to α(A) = 1. In that paper the masa A was shown to be
singular if α(A) > 0. Theorem 2.3 implies that a masa in a separable II1 factor is
singular if, and only if, it is strongly singular. This is the analogous result to Popa’s
δ–invariant one with unitaries in N replacing his nilpotent partial isometries, whose
domainsandrangesareorthogonalprojectionsinthemasa. ForamasaAinaseparable
II1 factor the results in [8] show that δ(A) is either 0 or 1, and Theorem 2.3 implies
that α(A) also takes only these two values (see [13] for the definition of α(A)).
The main result of the paper is Theorem 2.3, the proof of which is given in the
lemmas that precede it. These in turn are based on results of Sorin Popa, [10, Thm.
2.1, Cor. 2.3], which have their origin in [9]. We also give two applications of our
2
results. Oneshows thesingularity of tensor productsof singular masas (Corollary 2.4),
while the other shows that singular masas can usually be studied in the setting of
separable algebras (Theorem 2.5).
Much of the work in this paper was accomplished at the Spring Institute on Non-
commutative Geometry and Operator Algebras, held May 9–20 2005 at Vanderbilt
University. The lecture series presented by Sorin Popa at this conference provided the
basis for our results below. It is a pleasure to express our gratitude to the organizers
and to the NSF for providing financial support during the conference.
3
2 Main Results
Our first lemma is essentially contained in [10, Corollary 2.3]. The proof will amount
to identifying the subalgebras to which this corollary is applied.
Lemma 2.1. Let A be a masa in a II1 factor N and let e,f ∈ A be nonzero projections
with the property that no nonzero partial isometry w ∈ N satisfies the conditions
ww∗,ww∗ ∈ A, ww∗ ≤ e, w∗w ≤ f, and w∗Aw =Aw∗w. (2.1)
If ε > 0 and x1,...,xk ∈ N are given, then there exists a unitary u ∈ A such that
kEA(fxieux∗jf)k2 < ε (2.2)
for 1 ≤ i,j ≤ k.
Proof. Define two subalgebras B0 = Ae and B = Af of N. The hypothesis implies
the negation of the fourth condition for B0 and B in [10, Theorem 2.1] and so [10,
Corollary 2.3] can be applied. Thus, given a1,...,ak ∈ N and ε > 0, there exists a
unitary u1 ∈B0 such that
kEB(aiu1a∗j)k2 < ε, 1 ≤ i,j ≤ k. (2.3)
Theresultfollows from this by taking ai = fxie, 1 ≤ i≤ k, replacing u1 by theunitary
u= u1+(1−e) ∈A, and replacing EB in (2.3) by EA. Note that these two conditional
expectations agree on fNf.
Below, we will use the notation U(M) for the unitary group of any von Neumann
algebraM. Wewillalsoneedthewellknownfactthatifx∈ M andB isavonNeumann
subalgebra, then EB′∩M(x) is the unique element of minimal norm in the k·k2–closure
of conv{uxu∗: u ∈ U(B)}. We donothave anexact referencefor this, butitis implicit
in [1].
Lemma 2.2. Let A be a singular masa in a II1 factor N. If x1,...,xk ∈ N and ε > 0
are given, then there is a unitary u∈ A such that
kEA(xiux∗j)−EA(xi)uEA(x∗j)k2 < ε (2.4)
for 1 ≤ i,j ≤ k.
Proof. If x,y ∈ N and u ∈A, then
E (xuy∗)−E (x)uE (y∗)= E ((x−E (x))uE (y−E (y))∗) (2.5)
A A A A A A A
by the module properties of E . Thus (2.4) follows if we can establish that
A
kEA(xiux∗j)k2 < ε, 1≤ i,j ≤ k, (2.6)
4
when the x ’s also satisfy E (x ) = 0 for 1 ≤ i ≤ k. We assume this extra condition,
i A i
and prove (2.6). By scaling, there is no loss of generality in assuming kx k ≤ 1 for
i
1 ≤ i≤ k.
Let δ = ε/4. In the separable case, [7] gives a finite dimensional abelian subalgebra
A1 ⊆ A with minimal projections e1,...,en such that
kEA′∩N(xi)−EA(xi)k2 < δ (2.7)
1
for 1 ≤ i≤ k. The assumption that E (x )= 0 allows us to rewrite (2.7) as
A i
kEA′∩N(xi)k2 = kEA(xi)−EA′∩N(xi)k2 < δ (2.8)
1 1
for 1 ≤ i≤ k, leading to
n
k X emxiemk2 = kEA′∩N(xi)k2 < δ (2.9)
1
m=1
sincePnm=1emxiem = EA′1∩N(xi). Any partialisometryv ∈ N satisfyingvAv∗ = Avv∗
has the form pu for a projection p ∈ A and a normalizing unitary u ∈ N, [4, 5].
The singularity of A then shows that v ∈ A, making it impossible to satisfy the two
inequalitiesvv∗ ≤ e andv∗v ≤ (1−e )simultaneouslyunlessv = 0. Thusnononzero
m m
partial isometry v ∈ N satisfies vv∗ ≤ e , v∗v ≤ (1−e ), and vAv∗ = Avv∗. The
m m
hypothesis of Lemma 2.1 is satisfied, and applying this result with ε replaced by δ/n
gives unitaries u ∈ A such that
m
kEA((1−em)xiemumx∗j(1−em))k2 < δ/n (2.10)
n
for 1 ≤ m ≤ n and 1 ≤ i,j ≤ k. Define a unitary u ∈ A by u = u e , and let
Pm=1 m m
n
y = (1−e )x e , for 1 ≤ i ≤ k. We have
i Pm=1 m i m
n n
xi−yi = xi− X(1−em)xiem = X emxiem = EA′∩N(xi) (2.11)
1
m=1 m=1
for 1 ≤ i≤ k. The inequalities
kxi−yik2 < δ, kxi−yik ≤ kxik ≤ 1, kyik≤ 2, (2.12)
for 1 ≤ i≤ k, follow immediately from (2.8) and (2.11).
If we apply E to the identity
A
x ux∗ = (x −y )ux∗+y u(x∗−y∗)+y uy∗, (2.13)
i j i i j i j j i j
then (2.12) gives
kEA(xiux∗j)k2 ≤ kxi−yik2+kyikkxj −yjk2+kEA(yiuyj∗)k2
< 3δ+kEA(yiuyj∗)k2, (2.14)
5
and we estimate the last term. The identity
n n
yiuyj∗ = X (1−em)xiemuesx∗j(1−es) = X(1−em)xiemumx∗j(1−em) (2.15)
m,s=1 m=1
holds because each e commutes with u and e e = 0 for m 6= s. The last sum has n
s m s
terms, so the inequalities
kEA(yiuyj∗)k2 < δ, 1 ≤ i,j ≤ k, (2.16)
are immediate from (2.10). Together (2.14) and (2.16) yield
kEA(xiux∗j)k2 < 3δ+δ = ε, 1 ≤ i,j ≤ k, (2.17)
as required.
In the general case, we obtain A1 and (2.7) as follows. Since A is a masa,
EA′∩N(xi) = EA(xi)= 0, 1 ≤ i≤ k. (2.18)
Now EA′∩N(xi) is the element of minimal k·k2–norm in the k·k2–closed convex hull
of {wxiw∗: w ∈ U(A)}, so we may select a finite number of unitaries w1,...,wr ∈ A
such that each set Ω = conv{w x w∗: 1 ≤ j ≤ r}, 1 ≤ i ≤ k, contains an element
i j i j
whose k·k2–norm is less than δ. The spectral theorem allows us to make the further
assumption that each w has finite spectrum, whereupon these unitaries generate a
j
finite dimensional subalgebra A1 ⊆ A. Then, for 1 ≤ i ≤ k, EA′∩N(xi) is the element
1
of smallest norm in the k·k2–closed convex hull of {wxiw∗: w ∈U(A1)}, and since this
set contains Ω , we see that (2.7) is valid in general.
i
In [13], a masa A in a II1 factor N was defined to have the asymptotic homomor-
phism property (AHP) if there exists a unitary v ∈ A such that
lim kEA(xvny)−EA(x)vnEA(y)k2 = 0 (2.19)
|n|→∞
for all x,y ∈ N. In that paper it was shown that strong singularity is a consequence of
this property. Subsequentlyitwas observed in [12, Lemma2.1] thata weaker property,
which we will call the weak asymptotic homomorphism property, (WAHP), suffices to
imply strong singularity: given ε > 0 and x1,...,xk,y1,...,yk ∈ N, there exists a
unitary u∈ A such that
kEA(xiuyj)−EA(xi)uEA(yj)k2 < ε (2.20)
for 1 ≤ i,j ≤ k. Since the WAHP is a consequence of applying Lemma 2.2 to the set
of elements x1,...,xk,y1∗,...,yk∗ ∈ N, we immediately obtain the main result of the
paper from these remarks:
6
Theorem 2.3. Let A be a singular masa in a II1 factor N. Then A has the WAHP
and is strongly singular.
The following observation on the tensor product of masas may be known to some
experts, but we have not found a reference.
Corollary 2.4. For i = 1,2, let Ai ⊆ Ni be masas in II1 factors. If A1 and A2 are
both singular, then A1⊗A2 is also a singular masa in N1⊗N2.
Proof. Lemma 2.2 and the remarks preceding Theorem 2.3 show that singularity and
the WAHP are equivalent for masas in II1 factors. By Tomita’s commutant theorem,
A1⊗A2 is a masa in N1⊗N2, and it is straightforward to verify that the WAHP carries
over to tensor products (see [16, Proposition 1.4.27]), since it suffices to check this
property on a k·k2–norm dense set of elements, in this case the span of {x⊗y: x ∈
N1, y ∈ N2}.
As an application of these results, we end by showing that the study of singular
masas can, in many instances, be reduced to the separable case. The techniques of the
proof have their origin in [2, Section 7].
Theorem 2.5. Let N be a II1 factor with a singular masa A and let M0 be a separable
von Neumann subalgebra of N. Then there exists a separable subfactor M such that
M0 ⊆ M ⊆ N and M ∩A is a singular masa in M.
Proof. We will construct M as the weak closure of the union of an increasing sequence
M0 ⊆ M1 ⊆ M2 ⊆ ... of separable von Neumann subalgebras, chosen by induction.
These will have an increasing sequence of abelian subalgebras B ⊆ M with certain
k k
properties.
For a von Neumann algebra Q ⊆ N and for any x ∈ N, K (x) will denote the
Q
set conv{uxu∗: u ∈ U(Q)}, and the k·k– and k·k2–closures will be denoted KQn(x)
and Kw(x) respectively. The inclusions K (x) ⊆ Kn(x) ⊆ Kw(x) are immediate. The
Q Q Q Q
induction hypothesis is: each Mk is separable, Mk ⊆ Mk+1, and for a fixed sequence
{yk,r}∞r=1 in the unit ball of Mk which is k·k2–dense in the k·k2–closure of this ball,
(i) EA(yk,r) ∈ Bk+1∩KBwk+1(yk,r) for r ≥ 1, where Bk+1 = Mk+1∩A;
(ii) Kn (y )∩C1 is nonempty for r ≥ 1;
Mk+1 k,r
(iii) given ε >0, r ≥ 1 and a projection p ∈Bk, there exists u∈ U(Bk+1) such that
kEA((1−p)yk,spuyk∗,t(1−p))k2 < ε
for all 1 ≤ s,t ≤ r.
7
We first show that such a sequence of algebras leads to the desired conclusion. Let
M and B be respectively the weak closures of the unions of the M ’s and B ’s. Since
k k
Kn (x) ⊆ Kn (x) for all x∈ M and k ≥ 1, condition (ii) and a simple approximation
Mk M
argument show that Kw(x) contains a scalar operator for all x in the unit ball of M,
M
and thus for all x∈ M by scaling. Since Kw(z) = {z} for any central element z ∈ M,
M
this shows that M is a factor, separable by construction.
Now consider x ∈ M; scaling allows us to assume without loss of generality that
kxk ≤ 1. Condition (i) shows that E (y ) ∈ Kw(y ) for k,r ≥ 1, and an approxima-
A k,r B k,r
tion argument then shows that E (x) ∈ Kw(x) ⊆ Kw(x). Since E (x) is the element
A B A A
of minimal k·k2–norm in KAw(x), it also has this property in KBw(x). But this minimal
element is EB′∩M(x), showing that EB′∩M(x) = EA(x). If we further suppose that
x ∈ B′ ∩M, then x = E (x). Thus B′ ∩M ⊆ A and is abelian. Condition (i) also
A
shows that E (x) ∈ B, and so B′ ∩ M ⊆ B. Since B is abelian, we have equality,
A
proving that B is a masa in M. We can now conclude that E (y) = E (y) for all
A B
y ∈ M, and that B = M ∩A.
Another k·k2–approximation argument, starting from (iii), gives
inf{ max kEB((1−p)xipux∗j(1−p))k2: u∈ U(B)} = 0 (2.21)
1≤i,j≤r
for an arbitrary finite set of elements x1,...,xr ∈ M, kxik ≤ 1, and any projection p ∈
B, noting that E and E agree on M. By scaling, it is clear that this equation holds
A B
generally without the restriction kx k≤ 1. We have now established the conclusion of
i
Lemma 2.1 for B from which singularity of B follows, as in the proof of Lemma 2.2.
It remains to construct the appropriate subalgebras M .
k
To begin the induction, let B0 = M0 ∩A, and suppose that Bk ⊆ Mk have been
constructed. Consider a fixed sequence {yk,r}∞r=1 in the unit ball of Mk, k·k2–dense in
the k·k2–closure of this ball. The Dixmier approximation theorem, [6, Theorem 8.3.5],
allows us to obtain a countable number of unitaries, generating a separable subalgebra
Q0 ⊆ N, so that KQn0(yk,r)∩C1 is nonempty for r ≥ 1. Let {pm}∞m=1 be a sequence
which is k·k2–dense in the set of all projections in Bk. The singularity of A ensures
that the hypothesis of Lemma 2.1 is met when e = p and f = 1− p . Thus, for
m m
integers m,r,s ≥ 1, there is a unitary u ∈ A such that
m,r,s
kEA((1−pm)yk,ipmum,r,syk∗,j(1−pm))k2 < 1/s (2.22)
for 1 ≤ i,j ≤ r. These unitaries generate a separable von Neumann algebra Q1 ⊆ A,
and an approximation argument establishes (iii) provided that Q1 ⊆ Bk+1.
Since EA(yk,r) is the minimal k · k2–norm element in KAw(yk,r), we may find a
countable number of unitaries generating a separable subalgebra Q2 ⊆ A so that
EA(yk,r) ∈ KQw2(yk,r) for r ≥ 1. The proof is completed by letting Mk+1 be the
8
separable von Neumann algebra generated by Mk, EA(Mk), Q0, Q1 and Q2. The sub-
spaces EA(Mk) and Q2 are included to ensure that (i) is satisfied, (ii) holds by the
choice of Q0, and Q1 guarantees the validity of (iii).
9
References
[1] E. Christensen, Subalgebras of a finite algebra, Math. Ann., 243 (1979), 17–29.
[2] E. Christensen, F. Pop, A.M. Sinclair and R.R. Smith, Hochschild cohomology of
factors with property Γ, Ann. of Math., 158 (2003), 597-621.
[3] J. Dixmier, Sous-anneaux ab´eliens maximaux dans les facteurs de type fini, Ann.
Math., 59 (1954), 279–286.
[4] H. Dye, On groups of measure preserving transformations II, Amer. J. Math., 85
(1963), 551–576.
[5] V. Jones and S. Popa, Some properties of MASAs in factors. Invariant subspaces
and other topics (Timi¸soara/Herculane, 1981), pp. 89–102, Operator Theory:
Adv. Appl., by Birkh¨auser, Boston, 1982.
[6] R.V.KadisonandJ.R.Ringrose,Fundamentals of the theory of operator algebras,
Vol. II, Academic Press, Orlando, 1986.
[7] S. Popa, Singular maximal abelian *-subalgebras in continuous von Neumann
algebras, J. Funct.Anal., 50 (1983), 155–166.
[8] S.Popa, Onthedistance between masas intypeII1 factors, Mathematical physics
in mathematics and physics (Siena, 2000), 321–324, Fields Inst. Commun., 30,
Amer. Math. Soc., Providence, RI, 2001.
[9] S. Popa, On a class of type II1 factors with Betti numbers invariants, Ann.Math.,
to appear.
[10] S. Popa, Strong rigidity of II1 factors arising from malleable actions of w–rigid
groups, I, preprint, UCLA, 2003.
[11] S. Popa, A.M. Sinclair and R.R. Smith, Perturbations of subalgebras of type
II1 factors, J. Funct. Anal., 213 (2004), 346–379. (See Mathematics ArXiv
math.OA/0305444 for a corrected version)
[12] G. Robertson, A.M. Sinclair and R.R. Smith, Strong singularity for subalgebras
of finite factors, Internat. J. Math., 14 (2003), 235-258.
[13] A.M. Sinclair and R.R. Smith, Strongly singular masas in type II1 factors, Geom.
and Funct. Anal., 12 (2002), 199-216.
[14] A.M. Sinclair and R.R. Smith, The Laplacian masa in a free group factor, Trans.
A.M.S., 355 (2003), 465-475.
[15] S. A. White, Tauer masas in the hyperfinite II1 factor, Quart. J. Math. Oxford
Ser. (2), to appear.
[16] S. A. White, Tauer masas in the hyperfinite II1 factor, Ph.D. thesis, University of
Edinburgh, 2005.
10